fix diloco 2 #13
Merged
fix diloco 2 #13
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
samsja
force-pushed
the
fix-diloco-2
branch
from
e5fc6a0 to
294c36d
Compare
samsja
commented
Sep 27, 2024
JohannesHa
reviewed
Sep 27, 2024
Jackmin801
reviewed
Sep 27, 2024
src/zeroband/diloco.py
Outdated
| """ | ||
|
|
||
| self._logger.debug("sync inner model") | ||
| for param_offloaded, param in zip(self.param_list_cpu, model.parameters()): | ||
| param.data = param_offloaded.data.to("cuda") # todo: use copy_ here | ||
| param.data.copy_(param_offloaded.data.to(param.device)) # todo: use copy_ here |
There was a problem hiding this comment.
the .to(device) isnt needed here. You should be able to directly copy the data
There was a problem hiding this comment.
interesting so it will handle the move to device by itself ?
There was a problem hiding this comment.
my understanding is that a.copy_(b) means copy the content of b to a inplace so it works across devices and doesnt allocate a new intermediate storage.
meanwhile param.copy_(param_offloaded.data.to(param.device)) will first evaluate the .to which creates a new intermediate storage then copies the content. might be wrong tho, not sure how smart compilers/interpreters are
samsja
force-pushed
the
fix-diloco-2
branch
from
330bdd3 to
9d801d3
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
FIX DILOCO LFG
This pr fix the diloco intregration. The main bug was that model were not initialized the same across diloco rank. We used to fix this by pulling the weight init from hf. We don't do this anymore because we don't use transformer for the modeling.
This pr introduces a common seed. I checked, and the weights are correctly initialized now.
The loss is looking better now
Additionally the PR add testing for the diloco all reduce as well as some small fix and enhancement